💥 32 people on Ollama + Qwen3.8-27B. First token waits 102s. Default MTP draft… · @QwenAIFans · TwKit

@QwenAIFans

💥 32 people on Ollama + Qwen3.8-27B. First token waits 102s. Default MTP draft hits 76.2 tok/s for one user on a 4090. For a team it still queues one request at a time on this model. vLLM and llama.cpp kept batching. Solo box or shared? via ComputingForGeeks 👇 https://computingforgeeks.com/ollama-vs-vllm-vs-llama-cpp/

曝光 139 · 评论 2 · 点赞 0 · 书签 1